Author: Chris Carson
Date: 10:59:42 06/22/99
Go up one level in this thread
On June 22, 1999 at 11:08:17, Robert Hyatt wrote: >On June 22, 1999 at 10:20:27, Dann Corbit wrote: > >>On June 22, 1999 at 09:48:28, Dann Corbit wrote: >>[snip] >>>>The fact remains that computers have beaten GM's Thus they must be GM level to >>>>do it. Its like saying if one GM beats another that maybe that GM is not of the >>>>level, cause he only wins when a better GM blunders, get real !!! >>>This is not a demonstration of computers being at GM level. I have beaten >>>players who are *much* better than I am. It is not a demonstration that I am as >>>good as they are. It is merely a demonstration of an isolated win. >>I should mention also that beating them was not any sort of indication that I >>was inferior to them (even though I am). >> >>>>You are right we do not need an opinion poll question, they are GM level, >>>>otherwise they would not be able to beat a GM. Just because a GM blunders does >>>>not make him a GM anymore. Hoe many more win do computers have to do to make >>>>then GM level >>>Scientific proof is what is needed. Not an opinion poll. A win against a good >>>opponent does not prove equality. >>Computers *might* be at GM level. Or not. For a GM to be at GM level, what >>does he/she have to do? A computer must pass those exact same conditions or it >>is not *proven* to be at GM level. Period. Right now, we just don't know. >>Scientifically, that is. >> >>Let's invent a new measure called "Seems Like a GM to me" >>Any computer is at that level if you think it is. > > >Right now we are at computers 3, humans 5, in our 8 game 40/2hr series of >games. That _might_ mean the computers are at the lower GM level. It also >might mean that they are at super-GM level. Or it might mean they were somewhat >lucky. Untill we have enough games, we don't know. If we had a score of 15-5, >I think the conclusion would be pretty accurate (assuming 15 for humans) that >the computers are 200 points worse (ie 2400). If we had 10-10, I'd think that >we would conclude that the computers were reasonably close to 2600, although >there is still a significant margin of error for only 20 games. > >Or we could have a vote. That will decide it, right? :) Good point as usual Bob! :) The opinion poll question has no validity for actual computer strength, unless maybe the GM's that played the computers were the sample for the poll. I do think the poll will measure the general feeling this group has about the level of play of the computers. My guess is it is running high right now, but will be running lower when the next GM beats one of the top 10 in a 40/2hr or less match, higher when the reverse happens. :) My opinion is that the computers played well, but the GM's will adapt quicker than the released S/W (beta or development may improve faster). I think we are in a window where we will see some good results for both the computers and the GM's if the GM's can be paid to play (they are professionals). Was not that long ago that Rebel (very good program) lost to Rhode (hope my spelling is correct). It is my wish that computers could enter GM events (with the understanding that they do not share prizes), this would ensure GM's get paid and that programs could establish ratings against the GM's. Little chance that will happen. :) One human trait I am working on is adaptability for a chess program. I love speed, strong openings, learning, endgame tablebases, but I think adaptability would add something to todays programs, learning is a start and Crafty has opponent recognition (part of the adaptability puzzel). Perhaps preparation and some type of selective db knowledge (openings, learning and stearing in the middle/end game) to take advantage of an opponents weakness (based on games in a db or observed by the program base on games already played in a one on one match). Not sure, may be off base. I have some ideas here, will post at a later date. Best Regards, Chris Carson
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.